归一化算法实现畸变不变图像识别研究

Distortion-invariant image processing with standardization method

在光电混合联合变换相关器图像识别系统中,对与参考图像完全相同的真目标图像和不同的假目标图像能实现完全识别,但对于与参考图像间存在旋转和比例变化的目标图像却很难识别,甚至无法识别。由此,提出了基于图像质心和最长边的归一化图像处理新算法:根据图像像素总值求取图像质心,根据质心和图像边缘轮廓点之间的长度求出图像的最长边和旋转角度,然后根据统一标准将目标图像和参考图像调整至同一长度和旋转方向的完全相同的图像。最后给出了模拟和实验的相关结果,结果表明,该方法能实现存在旋转和比例变化的图像的畸变不变处理,从而提高了系统的识别能力。

In the image recognition system of an optical-electrical hybrid joint transform correlator, the same object and the false object comparing with reference image could be well recognized, but the distorted image of the scale and rotation transformation was difficultly recognized and even could not be recognized. A new standardization method is proposed to solve the problem of the distorted image recognition, which is that the image＇s centroid was gotten based on the pixels total value of the whole image, and then the largest apothem and the rotated angle were obtained by the distance of the centroid and the image brim, last the object and the reference image adjusted the entirely same images in the scale and the rotated angle according to the uniform standardization. Simulative and experimental results show that this image-processed method can realize the distortion-invariant processing of the scaled and rotated image and improve the recognized capability of optical-electrical hybrid correlator system.